AI / LLMChina
← Sticker Wall
JOURNEY · DETAIL
MetaS: The Difficult Pivot from AI Writing Assistant to AI Search
Founded: Yan Lipeng, Li Xiyao, Wang Hu · Shanghai MetaS Network Technology Co., Ltd.
JOURNEY
Key Fields
FIELD STAMPSIndustryAI / LLM
RegionChina
ScaleMid-size
ChannelOther
Origin
In 2018, while developing legal translation and contract review tools, the team discovered that Chinese lawyers and legal professionals were extremely inefficient at document proofreading and grammar correction. At the time, Grammarly was thriving globally, but there was no Chinese equivalent. In 2019, Yan Lipeng led the team to launch 'MetaS Writing Cat' (秘塔写作猫), a proofreading and polishing assistant for Chinese writing, with the initial goal of solving deep-seated inefficiencies in official document and legal drafting.
Milestones
2018
Cold Start Turning Point
The company was registered in Shanghai on January 2, 2018. The founding team, including Yan Lipeng and Li Xiyao, initially launched the MetaX legal translation tool for law firms and translation agencies. According to 36Kr, the team had fewer than 10 people at the time. Revenue came primarily from subscription fees for legal contract and document translation, totaling less than 1 million RMB annually, with tight cash flow and fragmented customers making scaling difficult.
2019
MVP PMF
In September 2019, the web version of 'MetaS Writing Cat' was officially launched, positioned as a Chinese grammar correction and polishing assistant. It attracted a large number of students and copywriters with its free correction features. By 2020, the user base exceeded 100,000, making it a rare word-of-mouth product in the domestic Chinese writing assistance track, though the paid conversion rate remained below 3%, making it difficult for the team to sustain itself.
2021
Financing & Expansion Growth
In April 2021, the company completed a Pre-A round of financing led by Vision Plus Capital and JOYY Inc., totaling tens of millions of RMB. The team expanded to 40 people, and paid membership and enterprise API versions were launched. According to official disclosures, annual revenue was approximately 5 million RMB, with paid users concentrated among university students and administrative staff, where low average revenue per user (ARPU) severely compressed profit margins.
2022
Model Impact Failure
Following the release of ChatGPT at the end of 2022, a large number of users shifted to general-purpose large models for writing and polishing tasks. MetaS Writing Cat's daily active users (DAU) declined continuously. According to 36Kr, the company lost nearly 40% of its paid users in the first half of 2023. The team attempted to launch full-process AI writing features, but the results fell short of expectations, and cash flow pressure surged. Yan Lipeng admitted that Grammarly would eventually be replaced by ChatGPT.
2023
Pivot & Trial Turning Point
In the second half of 2023, the team cut some features of Writing Cat and pivoted to developing the generative AI search product 'MetaS AI Search'. It was officially launched in March 2024, focusing on an ad-free, direct-answer search experience. It gained millions of visits in its first month, and according to SimilarWeb data, monthly visits once exceeded 10 million, making it one of the first domestic ad-free AI search products.
2024
Capital Recognition Growth
In 2024, the company completed its Series A financing, led by BlueRun Ventures, with participation from Vision Plus Capital and Super Symmetry Fund. According to 36Kr, the amount was in the tens of millions of USD. The company launched an industry-specific search version and expanded the team to over 80 people. However, it has yet to achieve profitability, with server and inference costs exceeding millions of RMB per month, keeping cash flow under pressure.
2025
Intensified Competition Turning Point
In 2025, with the successive launches of Kimi, Zhipu, and Chinese alternatives to Perplexity, general-purpose large model vendors flooded the AI search track. MetaS was forced to refocus on legal and academic vertical search. According to LatePost, annual revenue for the year was approximately 15 million RMB, still relying on financing to maintain operations. The team size shrank to about 60 people, forcing the company to repeatedly evaluate whether its differentiation barriers were solid enough.
Turning Points
- September 2019: Launched the web version of MetaS Writing Cat, finding its first PMF in the domestic Chinese writing assistance track by attracting students and copywriters with free features.
- End of 2022: After the release of ChatGPT, massive user churn occurred. Yan Lipeng admitted in a 36Kr interview that Grammarly would eventually be replaced by ChatGPT, forcing the team to rethink its technical roadmap.
- March 2024: MetaS AI Search was officially launched, focusing on an ad-free, direct-answer experience. It gained millions of visits in the first month, marking a complete transition from a writing assistant to a generative AI search product.
- 2025: As general-purpose large model vendors entered the AI search track, the company was forced to refocus on legal and academic vertical search, shrinking the team to 60 people and re-evaluating its competitive moats.
Failures & Pitfalls
- 2020: The paid conversion rate for Writing Cat remained below 3% for a long time, and the low ARPU from university students made it difficult to sustain the team, with early annual revenue under 1 million RMB.
- First half of 2023: The company lost nearly 40% of its paid users due to the impact of ChatGPT. Attempts to launch full-process AI writing features failed to meet expectations, leading to a sharp increase in cash flow pressure.
- 2025: Server and inference costs exceeded millions of RMB per month. Following the entry of general-purpose model vendors, the company was forced to downsize and remains unprofitable, continuing to rely on external financing to survive.
关键成功要素
- The product in the Writing Cat era was too dependent on a single correction feature and lacked deep workflow integration, making the cost for users to switch to ChatGPT almost zero.
- The choice of an ad-free model for AI search effectively acquired users but severely compressed monetization space, making it difficult to alleviate the pressure of marginal server and inference costs.
- The high barrier to entry in legal and academic fields became the core foundation that MetaS could defend, whereas writing users were the first to be fully cannibalized by general-purpose models.
- The team's choice to maintain a disciplined financing pace rather than burning cash aggressively, along with cautious cash flow management, was the key reason the company survived long enough to pivot.
Lessons
- If a vertical AI tool is merely a wrapper for general-purpose model capabilities, it quickly loses its value once large model vendors enter the same scenario.
- Users acquired through free services often have extremely low loyalty and will churn immediately when faced with stronger free alternatives; this type of user base is not a foundation for long-term survival.
- The AI search track appears to have a low barrier to entry but is actually extremely costly. The ad-free model requires a massive user scale to cover inference costs and support commercialization; small and medium-sized teams must constantly evaluate their viability.
- Maintaining a lean team and cautious cash flow management is a prerequisite for small and medium-sized AI startups to survive the wave of large models; blind expansion only accelerates failure.
Core Data
- Founded：2018
- Writing Cat Launch Date：September 2019
- 2020 User Count：Over 100,000
- Writing Cat Paid Conversion Rate：Below 3%
- 2021 Revenue：Approx. 5 million RMB
- Pre-A Financing：Tens of millions of RMB, April 2021
- Series A Financing：Tens of millions of USD, led by BlueRun Ventures in 2024
- 2025 Revenue：Approx. 15 million RMB
- 2025 Team Size：Approx. 60 people
- Monthly Inference Cost：Over millions of RMB
Competitors / Peers
In the domestic market, MetaS AI Search faces direct competition from Kimi (Moonshot AI), Zhipu Qingyan (Zhipu AI), and Quark AI Search. These three are deeply backed by large model vendors or internet giants, giving them significant advantages in inference costs and data resources. In the overseas market, it benchmarks against AI search platforms like Perplexity and You.com. Meanwhile, competitors from the MetaS Writing Cat era, such as Grammarly and WPS Lingxi, are now leveraging large model capabilities to continuously erode the Chinese writing assistance market, creating a severe competitive landscape for vertical tool entrepreneurs.